TokenRhythm releases NeoHorse-1 models trained on agent execution records

A post about NeoHorse-1 describes a training loop built around TokenRhythm’s OpenSquilla system. It captures agents’ model-routing decisions, tool calls, failures, recoveries and outcomes for training. The updated model then returns to OpenSquilla to generate the next round of records. The author describes this as early engineering validation of recursive self-improvement: feeding execution experience back into training so the next model can perform better.